When using native tool calling, the assistant's tool_calls must be saved in
message history. Without this, the model doesn't know it already called a
tool and may loop indefinitely calling the same tool.
This fix adds tool_calls forwarding in chatStreamResponse and chatWithOptions
history formatting.
- Add IOllamaTool and IOllamaToolCall types for native function calling
- Add think parameter to IOllamaModelOptions for reasoning models (GPT-OSS, QwQ)
- Add tools parameter to IOllamaChatOptions
- Add toolCalls to response interfaces (IOllamaStreamChunk, IOllamaChatResponse)
- Update chat(), chatStreamResponse(), collectStreamResponse(), chatWithOptions() to support native tools
- Parse tool_calls from Ollama API responses
- Add support for tool message role in conversation history
- Add StreamingChatOptions interface with onToken callback
- Add optional chatStreaming method to MultiModalModel abstract class
- Implement chatStreaming in OllamaProvider using collectStreamResponse
- Add IOllamaModelOptions interface for runtime options (num_ctx, temperature, etc.)
- Extend IOllamaProviderOptions with defaultOptions and defaultTimeout
- Add IOllamaChatOptions for per-request overrides
- Add IOllamaStreamChunk and IOllamaChatResponse interfaces
- Add chatStreamResponse() for async iteration with options
- Add collectStreamResponse() for streaming with progress callback
- Add chatWithOptions() for non-streaming with full options
- Update chat() to use defaultOptions and defaultTimeout
